Antelop
A database for systems neuroscience foundation models
Foundation models in neuroscience
- Large machine learning models that learn the relationship between neural activity, sensory inputs and motor outputs
- Trained via self-supervised learning methods such as masked modelling
- Can be fine tuned to perform a number of prediction tasks
- Require extremely large datasets, well structured and preprocessed
Data engineering
- Increasingly large datasets within systems neuroscience (e.g. Neuropixels)
- Data storage needs to keep up with data engineering best practices
- Custom file formats/project structures are hard to parse
- Data preprocessing pipelines becoming increasingly complex and computationally expensive
- Custom preprocessing/analysis scripts are very difficult to reproduce
- High entry barrier to existing tools like DataJoint and NWB which makes their adoption difficult for many labs
Our solution: Antelop
- Software package designed to facilitate the easy adoption of data processing and storage best practices
- Simple pip install and straightforward graphical configuration
- Extensive graphical user interface for all aspects of your data management and processing
- MySQL database backend for centralised storage
- Supports electrophysiology, calcium imaging, and behavioural data processing with HPC integration
Our solution: Antelop
- Integrates with existing tools, such as popular spikesorters, CaImAn, and DeepLabCut
- Implements a range of data visualisation tools and metrics out of the box, including an analysis standard library
- Supports the writing of custom analysis scripts, with direct integration to your lab’s GitHub and data immutability checks for reproducibility
- Has import/export functions for NWB and a range of acquisition systems
- Has an opinionated but accomodating database structure for ML models to utilise
Publishing
- Working on a preprint at present
- We aim to publish by May this year
- Python package has been released but is undergoing extensive testing